🧠 SHL Product Recommendation API

This API provides intelligent recommendations for SHL assessments based on user queries using a Retrieval-Augmented Generation (RAG) approach. It uses semantic search over vector embeddings generated from SHL product descriptions.

---

🚀 Live Demos

• 🔍 Streamlit Demo App (User-facing search interface): https://shl-troger.streamlit.app/
• 🔗 API Endpoint (FastAPI): https://shl-api-ut7j.onrender.com
  • Interactive docs: https://shl-api-ut7j.onrender.com/docs

---

🧰 Tech Stack

• FastAPI – for backend API.
• Streamlit – for frontend demo.
• Sentence Transformers – for embedding queries and SHL products.
• Supabase – as a vector database backend.
• Uvicorn – as the ASGI server.

---

📦 Installation (For Local Setup)

git clone https://github.com/your-username/shl-recommendation-api
cd shl-recommendation-api
pip install -r requirements.txt

---

▶️ Running Locally

Start API:

uvicorn main:app --reload

Start Streamlit Frontend:

streamlit run app.py

---

📨 API Usage

POST /recommend

• Request Body:

{
  "query": "Hiring for Python",
  "threshold": 0.4,
  "top_k": 10
}

• Response:

{
  "results": [
    {
      "id": "50e6231d-8f64-4b66-aa39-1ad193826151",
      "name": "Python (New)",
      "url": "https://www.shl.com/solutions/products/product-catalog/view/python-new/",
      "remote_testing": true,
      "adaptive_irt": false,
      "duration": "11 minutes",
      "test_types": "Knowledge & Skills",
      "similarity": 0.536763163612078
    }
  ]
}

✅ You can also test the API at: https://your-api-url/docs

---

📌 Notes

• The model used is all-MiniLM-L6-v2 from HuggingFace for fast and accurate sentence embeddings.
• SHL product embeddings are stored in Supabase with vector indexing and similarity scoring.
• Threshold filtering is applied to ensure relevance of results.

---

🧑‍💻 Authors

• Ayush Gupta

---